Voice Ledger

The Challenge

The recorders people actually reach for are built for video calls, and they meter minutes because cloud speech recognition bills by the minute. That pricing model has a hard edge: the popular consumer tier caps a single recording at ninety minutes. A student recording one three-hour seminar a week cannot use it at all, and finds that out partway through the seminar.

The second problem is quieter. A lecture, a client interview, a doctor’s appointment: these are recordings people would rather not hand to a server, and a cloud service cannot offer that at any price.

Our Approach

Voice Ledger does the recognition work on the device, which removes server metering and the app’s need to upload a recording. Apple may need to supply a recognition-language asset. Optional AI reports and questions use a Gemma or LFM model downloaded directly from Hugging Face only after the user starts the transfer. After the needed assets are installed, a basement lecture theatre, plane, or field interview does not need a transcription or AI server.

The transcript is treated as a draft rather than a verdict. Speech recognition gets things wrong, so every line is editable, and an edit keeps its link back to the audio: correct a garbled sentence and tapping it still plays the second it came from. That link is what makes an imperfect transcript useful instead of annoying.

Key Features

• No length limit. Record a three-hour lecture in one sitting; the app shows how many hours will fit before you start, at roughly 14 MB per hour.
• On-device transcription. Word-level timing after the required recognition assets are installed; recordings are not sent to Studio AM.
• An editable transcript. Fix any line; corrections stay anchored to the audio.
• Import what you already have. Audio and video files to transcribe, or SRT, VTT and plain-text transcripts with their timings.
• Markers and speaker names. One tap flags a moment while it is happening; speakers are named by hand and the name applies everywhere.
• Built for long sessions. The Lock Screen shows recording is still running, a warm device is told about honestly instead of stalling silently, and a recording cut short by a crash is recovered on the next launch.
• Export, never gated. Audio, plain text, SRT/VTT captions, or lossless JSON, with or without a subscription.

What It Deliberately Does Not Do

It does not separate speakers automatically. Apple’s on-device speech engine has no such feature, so shipping one would mean guessing, and a confidently wrong “Speaker 2” is worse than an honest blank.

It does not fetch media from YouTube or any other service. It works with files you already have and text you paste, a boundary written into the code, not just the copy.

Access and Data

Recording, transcription and importing require verified access, with no free tier. In the US, the monthly plan is $7.99 and may include a 3-day trial for eligible new subscribers. The yearly plan is $39.99 with no introductory offer. Apple shows current local terms and eligibility on the purchase screen. Everything already captured stays readable, playable, searchable and exportable under the app’s access rules.

There is no Studio AM account, advertising, Studio AM analytics, or tracking. Recordings and transcripts stay on the device and are never included in a model request. Network access is used for StoreKit, Apple-provided speech assets, and optional on-device AI models that download directly from Hugging Face only after the user starts the transfer. Hugging Face may process connection, diagnostic, and download-interaction metadata under its own policy; Studio AM receives none of that provider data.

Results

Built on StudioMemoKit, the same recording and transcription engine that ships inside Second Brain and Daily Vibe, pointed here at the case those apps were never meant to carry: the three-hour recording.

Common questions

Voice Ledger is an English-language Mac, iPhone, and iPad app for recording long talks and turning them into editable, time-aligned transcripts. It records lectures, meetings, and interviews with no app-imposed per-recording length limit, then links transcript words back to their audio positions. Users can import audio, video, SRT, VTT, or text files they own and export text, captions, audio, or JSON. Speakers are named manually rather than detected automatically. Recognition languages may require an Apple-provided download. Optional on-device AI features use a Gemma or LFM model that downloads directly from Hugging Face only after the user starts the transfer; the request contains no recording, audio, transcript, question, summary, or other saved app content, and Studio AM receives none of Hugging Face's provider data. After the needed assets are installed, processing stays on the device. Voice Ledger does not fetch media from a URL. There is no free tier: recording, transcription, importing, and editing require verified access. The US monthly plan is $7.99 and may include a 3-day trial for eligible new subscribers; the $39.99 yearly plan has no introductory offer. Existing logs remain readable, playable, searchable, exportable, and deletable under the app's access rules. There is no Studio AM account, advertising, Studio AM analytics, or tracking.

How long can one recording be?

There is no limit set by the app. A three-hour lecture is a normal recording: it transcribes in a few minutes and uses about 14 MB of storage per hour of audio. The practical limit is free space on the device, and the app shows how many hours will fit before you start.

Does it work without an internet connection?

Recording, editing, search and export are local. Apple may need to supply a recognition-language asset, and optional AI features require a user-started model download from Hugging Face. Once the assets you need are installed, transcription and AI processing run on the device. StoreKit also uses a connection for plans, purchase and restore.

Does my audio get uploaded anywhere?

No. Recordings and transcripts are stored on the device and are not transmitted to Studio AM. There is no account, no sign-in, and no server of ours involved at any point.

Can I correct mistakes in the transcript?

Yes. The transcript is a draft you can edit, and corrections keep their link to the audio: tapping an edited line still plays the moment it came from. You can also name speakers by hand, and a name you set applies everywhere it appears.

Does it separate speakers automatically?

No. Apple's on-device speech engine has no speaker separation, so the app does not claim to do it. You label speakers yourself, which is accurate because you decide it. A wrong "Speaker 2" is worse than none.

Can I bring in recordings or transcripts I already have?

Yes. Import audio and video files and transcribe them, or import a transcript as SRT, VTT or plain text with its timings preserved. The app never downloads or extracts media from YouTube or any other service. It works only with files you supply or text you paste.

What happens if my subscription ends?

Nothing is taken away. Everything already recorded or imported stays readable, playable, searchable and exportable. Export (audio, plain text, SRT/VTT captions, or lossless JSON) is never behind the subscription. The subscription covers making new recordings, running transcription, and importing.

How much does it cost? Is there a free tier?

There is no free tier. Recording, transcription and importing require a subscription. In the US, the monthly plan is $7.99 and may include a 3-day trial for eligible new subscribers. The yearly plan is $39.99 with no introductory offer. Apple shows eligibility and local terms before purchase.